audience preference and editorial judgment: a …...school of journalism & mass communication...
TRANSCRIPT
1
Audience Preference and Editorial Judgment: A Study of Time-Lagged Influence in Online News
Angela M. Lee, M.A. Doctoral Student
School of Journalism The University of Texas at Austin
Austin, TX 78712 Email: [email protected]
&
Seth C. Lewis, Ph.D. Assistant Professor
School of Journalism & Mass Communication University of Minnesota Minneapolis, MN 55446 Email: [email protected]
Lee, A. M. & Lewis, S. C. (2012, April 21). Audience preference and editorial judgment:
A study of time-lagged influence in online news. Paper presented at the 13th International Symposium on Online Journalism, Austin, TX.
2
Abstract The rise of sophisticated tools for tracking audiences online has begun to change the way media producers think about media audiences. This study examines this phenomenon in journalism. Research suggests that journalists, after long resisting or ignoring audience preferences, are becoming increasingly aware of user desires, manifest via metrics. However, research also finds a gap in the news preferences of journalists and audiences. This study asks: who influences whom more in this disparity? Through longitudinal secondary data analysis of three U.S. online newspapers, and using structural equation modeling, this study finds (1) audience preferences affect subsequent editorial placement of news stories, (2) such influence intensifies during the course of the day, (3) editorial judgment does not influence subsequent audience preference for news stories, and (4) audience preferences affect editorial judgments more than the other way around. Implications of these findings and suggestions for future research are discussed. Keywords: audiences, journalism, metrics, new media, online news
3
Audience Preference and Editorial Judgment: A Study of Time-Lagged Influence in Online News
Introduction
This paper addresses a question of growing relevance for journalistic practice and
the scholars who study it: As audiences and their content preferences are rendered
increasingly visible through digital tracking metrics and features such as “most viewed”
lists on news homepages, to what extent has this development affected editorial
decisions, if at all? An emerging body of literature (e.g., see Anderson, 2011a,b;
Boczkowski, 2010; Dick, 2011; Loosen & Schmidt, 2012; Lowrey, 2009; MacGregor,
2007) suggests that journalists, after decades of disregarding audience desires as
insignificant to the news process, are becoming more aware of and adaptive to consumer
tastes, even as they rationalize and reduce the audience to a quantifiable aggregate
(Napoli, 2010). Simultaneously, journalists are coming to recognize audiences’ growing
power to personalize their news experience (Thurman, 2011) and register their interests
in a way that shapes the news patterns of fellow users (Thorson, 2008); this sets forth an
“agenda of the audience” (Anderson, 2011b, p. 529) that challenges traditional notions of
the gatekeeping and agenda-setting functions of the press (Singer, 2011).
Amid this uncertainty about audience influence on professional news judgment,
Pablo Boczkowski and his colleagues have made a vital contribution to the literature with
very recent work (in particular, Boczkowski, 2010; Boczkowski, Mitchelstein, & Walter,
2011; Boczkowski & Peer, 2011) that has found a significant gap between the news
preferences of journalists and consumers: namely, that journalists generally prefer “hard”
4
news (public affairs) while consumers generally prefer “soft” news (non-public affairs).1
Missing from their analysis, however, is an examination of time-lagged influence one
way or another. That is, if journalists and audiences indeed want different kinds of
content online, who influences whom more in this taste disparity, if at all? If journalists
signal that something is important at Time A, are audiences more or less likely to
recognize that salience at Time B—and vice versa?
This paper takes up these questions. By examining editorial and consumer
preferences gathered from the websites of three major U.S. newspapers, we find that,
controlling for potential reciprocal effects, (1) audience preferences affect subsequent
editorial placement of news stories, based on time-lagged analysis; (2) the strength of
audience preferences’ effect on editorial judgments intensifies during the course of the
day; (3) there is no overall lagged effect of editorial judgments on audience preferences,
but subsequent analysis suggests this finding may be explained by the dynamic nature of
the lagged effects observed during the course of the day; and (4) the lagged effect of
audience preferences on editorial judgments is stronger than the lagged effect of editorial
judgments on audience preferences.
The article proceeds by exploring journalists’ conception of the audience, both
historically and in the present era of increasingly sophisticated metrics, then reviews
emerging research on audience influences in newswork. Thereafter, an elaboration on
Boczkowski and his colleagues’ work serves to introduce our hypotheses, research
questions, and secondary data analysis, followed by a results section that engages a
variety of longitudinal statistical models under the Structural Equation Modeling
1 For a thorough discussion of the distinctions between hard and soft news, see Reinemann et al. (2012), in addition to the aforementioned work of Boczkowski and colleagues.
5
framework to render a more thorough picture of this data. Implications for newswork and
future research on editorial and audience desires are discussed in the concluding section.
Literature Review
Journalists, Audiences, and Metrics
In a mid-sized daily newspaper in the American South, administrators have
mounted a giant board covered with rows of newspaper front pages from previous
weeks. Each page accompanies a chart with circulation and penetration figures for
that day. The board’s purpose is clear—it’s an oracle, offering cryptic truths about
the hidden connection between news content and audience behavior. And the
truths remain cryptic. As one editor puts it, “If anyone in this building tells you
they know how to interpret this, they’re full of it.” (Lowrey, 2009, p.44)
Lowrey’s anecdote serves to reinforce a baseline assumption about journalists’
news decisions relative to audience interests—namely, that they are (or have been) miles
apart. Historically, journalists have neither understood nor truly cared to understand their
audiences’ desires for certain varieties of news content, preferring instead to trust their
own professional instincts and the cues provided by peer institutions in making news
judgments (Lowrey, 2009; see related discussion in Ettema & Whitney, 1994; Fishman,
1980; Schlesinger, 1987; Schudson, 1992; Tuchman, 1978; Weaver & Wilhoit, 1996). In
perhaps the most important study of this disconnect, Gans (2004) found that journalists
disregarded both “qualitative” (letters to the editor) and “quantitative” (audience research
statistics) forms of feedback, instead preferring to craft content with their bosses and
themselves in mind, assuming that “what interested them would interest the audience”
(2004, p. 229; see discussion in Anderson, 2011b, p. 531). And yet, even as journalists
6
traditionally have “ignored, rationalized away, or redefined audiences to suit their own
needs or to conform to constraints” (Lowrey, 2009, p. 44), the above anecdote also
suggests that the status quo—like the fortunes of newspapers themselves—is changing, as
news organizations rely less on crude (analog) guesswork and become increasingly
plugged in to relatively precise (digital) tracking of audience behavior online.
For media work broadly, the emergence of increasingly sophisticated analytic
engines for gathering and assessing digital footprints—uniques, pageviews, time on site,
engagement, “likes,” etc.—has begun to transform the way media producers and their
advertisers think about audiences: their present and future behaviors, their context-driven
tastes, and their relative economic value (Napoli, 2010). The rise of these “audience
information systems” has led media workers across all sectors to envision audiences in
quantifiable, data-driven ways, thus privileging a scientific precision over vague
impressions (Napoli, 2010, p. 11). Of course, the “rationalization of audience
understanding” described by Napoli (p. 31)—this effort to bring empirical rigor and
quantitative methods to bear on the process of making sense of audience behavior—is not
merely an internet-era phenomenon (e.g., see Bogart, 1957), but is, in fact, part of a larger
and ongoing effort in the media industries to facilitate greater management of and control
over audiences by virtue of knowledge about them (c.f., Ettema & Whitney, 1994;
Turow, 2005). For media scholars, particularly those coming from a political economy
viewpoint, this “audience as product” perspective is central to the critique that, above all,
capitalistic media “manufacture” audiences to be sold as commodities to advertisers
(Smythe, 1977; c.f., Ang, 1991). In this view, audience measurement is not merely a
subsidiary spot-check of size and demographics, but represents a complex integration of
7
actors, technologies, and behaviors—“a set of measurement procedures that are shaped
by both industry dynamics and the technological and usage patterns of the media whose
audience is manufactured” (Bermejo, 2009, p. 138; see also Schmidt & Loosen, 2012). In
the shift from the broadcast era to the digital era, the tools used for tracking audiences
have changed (Bermejo, 2009), as have the types of audience “commodities” sold to
advertisers (Lee, 2011), but nevertheless the core function of rationalizing audiences
remains integral to media production and distribution processes. Likewise, market-driven
forms of journalism (Cohen, 2002; McManus, 1994) are by no means unique to the
internet era and the audience-tracking enhancements it presents. What is unique about the
present moment is the sheer volume of audience data, generated by the ease and ubiquity
of digital tracking technologies, as well as the extent to which that data can serve to
influence media work by more fully communicating audience preferences.
For newswork, this development would appear to challenge the “journalistic gut
feeling” typically associated with individual and collective decision-making about
newsworthiness (Schultz, 2007). A growing number of newsroom-based studies (most
notably Anderson, 2011a; Bozckowski, 2010; Dick, 2011; MacGregor, 2007) have shown
that journalists are becoming more conscious of and responsive to audience traffic
statistics, notwithstanding the frequent discrepancies and problems inherent in such data
(Graves & Kelly, 2010; see also Bermejo, 2009).
This management-driven emphasis on metrics has been interpreted differently by
journalists in different newsrooms (Usher, 2010). But in many cases there is an implicit,
if not outright, encouragement for search engine optimization (SEO) in the selection and
editing of news (Dick, 2011)—that is, an often subtle but sometimes deliberate pursuit of
8
topics and terminology most likely to attract traffic via search algorithms and viral social
channels. Usher (2010) describes the consternation this has caused for some journalists:
The demand that every story generate traffic creates, in their minds, horrible
pressure to produce work that will be measured only by how much it is read. The
more they can seed their work with SEO terms and then promote their work on
social media platforms, the better their metrics and the happier their bosses. But
that formula can make for some very exhausted journalists. (para. 3)
What’s striking about the present use of audience metrics in newswork is that the data are
increasingly specific (tracked story-by-story, user-by-user, and even journalist-by-
journalist; see Garber, 2011) and also more widely circulated among reporters and editors
(as opposed to business-side management alone), amplifying the effect of this metric
awareness (Anderson, 2011a). This contributes to “a culture of the click” kind of logic
that is beginning to permeate contemporary newswork, and journalists are struggling to
balance this new pursuit of pageviews with a desire to preserve traditional norms, values,
and professional judgment (Anderson, 2011a). The overall presumption is that audience-
tracking data are contributing to editorial decision-making like never before.
Audience Preferences and Power
Simultaneously, even as audiences are seen in an “algorithmic” fashion, reduced
to statistical flatness (Anderson, 2011b), the rise of user-generated content sites and
social media spaces has generated much discussion about the relative empowerment of
the “people formerly known as the audience” (Rosen, 2006; see also discussion in Bruns,
2008; Gillmor, 2004; Harrison & Barthel, 2009; Jenkins et al., forthcoming; Jönsson &
Örnebring, 2011; Lewis, Kaufhold, & Lasorsa, 2010; Mitchelstein & Boczkowski, 2009,
9
2010; Mortensen, 2011; Postigo, 2011; Robinson, 2011; Shirky, 2008; Singer et al., 2011;
Thurman, 2011, p. 397). Audiences, in this view, are assumed to be more autonomous in
personalizing their media, filtering the news, and otherwise disregarding the editorial
cues provided by journalists (see related discussion in Mersey, 2010, and Thurman,
2011)—exercising a selectivity of exposure that scholars suggested would become a
defining feature of 21st century media (Chaffee & Metzger, 2001).
This conception of the audience sees the audience neither as a group of people
appropriating media as recipients, nor being appropriated by media industries as
commodities, but rather as active participants in a digital network of communication,
collaborating in the creation and diffusion of information (Schmidt & Loosen, 2012). The
effect of such activity is the individual and collective manifestation of audience desire:
qualitatively in the texts produced through user-generated content across a range of
personal and public online spaces, or quantitatively in the metrics derived from tracking
how users share, “like,” and annotate media. For news, this occurs as audience choices
become more visible on “most popular” lists on news sites (Thorson, 2008), and as
audience-supplied content becomes more integrated into the news process, both via news
organization websites (Robinson, 2011; Singer et al., 2011) and social media spaces such
as Twitter (Hermida, 2010; Lasorsa, Lewis, & Holton, 2012). While it’s true that most
audience members neither want to participate in news production on a large scale nor are
allowed to do so meaningfully on most news sites (Domingo et al., 2008), nonetheless
they may contribute simply by clicking on a story and enhancing its popularity, signaling
a level of salience that both captures and communicates publicly an audience’s collective
interest.
10
This aggregated “agenda of the audience” (Anderson, 2011b, p. 529) is revealed
not only below the surface via tracking data but also—and perhaps more importantly—
above the surface, even on news homepages themselves, via interactive features such as
lists of “most viewed” or “most emailed” stories (Boczkowski & Mitchelstein, 2012).
Thorson (2008) refers to such lists as “news recommendation engines” that serve as a
way to access content, and an accumulation of actions taken around that content (p. 477).
Such lists offer a “public endorsement” (p. 475) that is neither entirely institutional nor
individual in nature, but nevertheless represents an aggregate wisdom of site visitors, and
therefore has a shaping influence on how certain articles are evaluated, internally (by
journalists) and externally (by audiences). These forms of public visibility and audience
endorsement are presumed to have a growing impact on editorial decisions as journalists
take agenda-setting cues from the popularity of certain stories or genres (Chaffee &
Metzger, 2001; c.f., McCombs, 2005).
Who is Influencing Whom?
Putting these two trends in concert—the concurrent rise of audience metrics and
audience influence—raises an important question taken up in this paper’s analysis: To
what extent do user preferences affect editorial decisions? Scholarly research is only
beginning to untangle this question.
In a survey of the British local press, Singer (2011) asked top editors to identify
their newspaper’s best online work from the previous year as well as their online content
that generated the most traffic. Her subsequent analysis found that only a third of the
items cited as sources of pride were also the most popular with audiences, though there
was some degree of “agenda overlap” on topics such as sports. Overall, and “despite
11
excruciatingly detailed ‘hit log’ data, online audience preferences do not seem to be
having a notable … agenda-setting impact on local editors” (Singer, 2011, 636). This
supports the general finding in journalism studies that, amid opportunities for audience
participation and inclusion in the news process, journalists keep at the margins those
outside influences that might reshape their values or practices (Singer et al., 2011).
However, Singer’s (2011) finding is limited by a research design that neither
captures ethnographic-rich complexity nor quantitative precision with regard to actual
editorial vs. audience preferences. More revealing is the limited but growing body of
evidence from newsroom-based ethnographies and interviews that have focused on
audience metrics. In his study of British online journalism, MacGregor (2007) found that
newsworkers, especially web editors, rely on audience tracking as handy feedback,
sometimes monitoring metrics “obsessively” and occasionally “re-weighting” editorial
priorities as a result. From some of his respondents, however, it became apparent that
journalists feared “any slide towards populism” and its associated “slavery” to audience
fancy, thus reinforcing the professional resistance described by Gans (2004) and
Schlesinger (1987) in their studies of pre-internet journalism. Overall, the most
prominent change MacGregor found was to the news “instinct” (c.f., Schultz, 2007):
“Some admit now that they double-check their instinctive guesses with tracking data.
They no longer implicitly trust themselves” (MacGregor, 2007, p. 294). A similar kind of
hesitation and unease is apparent in Boczkowski’s (2010) study of South American
newsrooms—and yet he found that journalists generally held their ground against the
encroachment of audience desires. Boczkowski notes that, as journalists become
increasingly aware of audience preferences for “softer” news in contrast to their
12
professional desires for hard news, they
experience a tension between the overall preferences revealed by these choices
and dominant occupational values, yet they stick to these values in the face of
dissonant consumer preferences. [Thus] … the logic of the occupation prevails
over the logic of the market… (p. 253)
This tension between what audiences want to read and what journalists think they
should read thus functions as the flash point around which questions of search-engine
optimizing, traffic-chasing, and editorial decision-making are negotiated among a range
of newsroom actors: managers, homepage editors, technology specialists, and newly
emerging “SEO teams.” The scant literature, to date, suggests that while organizations
are designing schema to optimize content for audiences (Dick, 2011), and individual
journalists are widely being encouraged to “connect” with audiences (Robinson, 2011b,
p. 1130), compressing old barriers of time and space between journalist and news
consumer (Weiss & de Macedo Higgins Joyce, 2009)—nevertheless, there remains an
underlying preservation of and perseverance for professional ideals (Singer, 2011). Dick
(2011) describes this impulse in his study of SEO practices at several UK news
organizations, from which he concludes:
The culture of SEO in these organisations is visible, but not dogmatic. This
research found only one example of a media organisation where SEO directs
editorial decision-making, for all others SEO is acknowledged to inform (by
varying degrees), but never lead editorial, with significant resistance to this idea in
some quarters. It is acknowledged that in yielding to editorial (and style)
conventions these organisations are losing out on traffic. But this is broadly
13
considered a price worth paying in order to preserve the brand, and maintain the
news values of the profession. (p. 474)
Just as Dick (2011) found that SEO strategy implementation varied by news organization,
Anderson (2011a,b), in his study of two large U.S. newsrooms, discovered that in one
journalists were “excited” about the “sudden and immediate glimpse into the mind of
their audience” made possible by metrics; this was facilitated by the ready circulation of
those metrics by managers eager to use them as strategic tools in shaping newsroom
practices (2011b, p. 536). In the other newsroom, meanwhile, metrics received less
attention from management and thus had less day-to-day influence on the ground-level
work of reporters (Anderson, 2011a).
Boczkowski’s earlier research (2004) has shown that technology is socially
shaped in its introduction to newsrooms—that environmental contexts, work patterns, and
organizational visions of technology, all intermingled with professional identity and
culture, contribute to the so-called “impact” that new tools have for newswork (see
Anderson, 2011a, p. 552). Likewise, literature on audience-tracking tools and their
incorporation in news settings, combined with anecdotal evidence in industry reports
(e.g., Usher, 2010), suggests that adoption varies across organizational settings,
depending on management directives and the manner in which they are communicated.
However, when considering these case studies in total, a composite picture begins to
emerge: The more journalists know about their audience metrics, the more they become
“sensitive to the implications of what their audience [is] reading and why,” altogether
showing that “the process of ‘deciding what’s news’ is increasingly influenced by
quantitative audience measurement techniques” (Anderson, 2011a, p. 563).
14
These case studies, while thick with description about newsroom culture and
practice, nevertheless are thin in providing empirical data from which to measure,
quantitatively, the extent to which audience behaviors might be influencing editorial
decisions. In very recent times, Boczkowski and his colleagues (see especially
Boczkowski et al., 2011; Boczkowski & Peer, 2011) have begun to address this gap by
calculating the relative congruence between journalists’ choices (signaled by story
placement on news homepages) and audiences’ preferences (indicated by “most viewed”
lists on the same news homepages). Their methods involved capturing data at a single
point in time (Boczkowski et al., 2011) or at several points during the day (Boczkowski
& Peer, 2001) and assessing this “snapshot” of thematic preferences, i.e. the relative
proportion of public affairs news represented among the top 10 choices for both the
journalist and audience groups. Their findings were generally significant and consistent,
indicating that there is a major gap between journalists’ and consumers’ news choices—
the former preferring “hard” news and the latter preferring “soft” news.
Boczkowski and colleagues thus found that journalists and audiences want
different things. But that leaves open the question: Who is influencing whom more in this
taste disparity, if at all? What’s missing in the literature is an examination of time-lagged
influence—in effect, a more thorough accounting for particular stories and their
longitudinal lifespan, either on the journalists’ top 10 list (as represented by story
placement), the audience’s top 10 list (as represented on “most viewed” lists), or both. If
journalists give prominence to a news story at Time 1, are audiences more or less likely
to have recognized that importance at Time 2, by making it a “most viewed” story? And
how does that process work in reverse, such that journalists take cues from audience
15
choices reflected in “most viewed” lists?
These questions point to the shifting nature of gatekeeping (Shoemaker & Vos,
2009) and agenda-setting (McCombs, 2005) in a digital environment that, by nature,
reconfigures traditional mass communication roles (Chaffee & Metzger, 2001). To the
extent that users take their cues from and are influenced by audience-driven features such
as “most viewed” lists—and there is good evidence to suggest that they do (Berger &
Milkman, forthcoming)—then whither the “human information filter” (Barzilai-Nahon,
2009) that historically has been the journalist? Moreover, to the extent that journalists
likewise take cues from audience desires—e.g., by adjusting the placement of stories to
correspond with the level of interest apparent on the “most viewed” list—then whither
the agenda-setting function of the press in this process? Clearly, journalists still set forth
an agenda and determine what passes through their gates at the point of initial
publication. And yet, what happens after that point—how news items generate a reaction
among audiences, and how such a reaction might affect later editorial judgment—remains
rather unclear and yet vital to explore. Singer (2011) identified this as a critical
deficiency in the study of journalists and audience metrics: the lack of a “longitudinal
analysis to determine the impact, if any, of user data or input over time” (p. 636).
This paper takes up that problem. Using fixed effects regression models to plot
time-lagged analysis, we seek to advance scholarly and professional understandings of
the causal relationship between technological quantification (e.g., audience metrics) and
human qualifications (e.g., editorial decisions) in dictating the future of journalistic
gatekeeping.
16
Hypotheses
While the relative lack of existing empirical studies prevents us from making
directional hypotheses (i.e., whether the lagged impact is positive or negative), the
evolution of online journalism and extant research do indicate the possibility of audience
preferences having a lagged influence on editorial decisions as newsroom editors
gradually base the placement of homepage news stories on aggregated popularity of
certain news items. Following this logic, we propose:
H1: Audience preferences (e.g., “most viewed” ranks) have a significant overall lagged effect on online editorial judgments (e.g., placement of news stories on homepages), controlling for potential reciprocal effects. RQ1: What is the nature of the lagged effect of audience preferences on editorial judgments?
Moreover, while time-lagged effects have not, to our knowledge, been examined
empirically, research seems to suggest that audience preferences are having some effect
on editorial judgments wherein they are driving story placements despite resistance on
the part of journalists, as Boczkowski’s recent work indicates. Such tension between what
journalists think audiences should read (e.g., public affairs topics) and what audiences
prefer reading (e.g., non-public affairs topics) paints a mixed picture at best regarding the
relationship between audience preferences and editorial judgments. Nonetheless, given
the cumulative nature of tracking devices today, such that as the day goes on newsrooms
will only have more data on audiences’ likes and dislikes as more people consume news
online, we hypothesize that the effect of audience rank on editorial rank will strengthen
over the course of an average day:
H2: The effect of audience preferences on editorial judgments will increase steadily across the four recorded time points on an average day.
17
Just as audience rank is likely to have lagged effect on editorial rank, and
extending the rationale from classic gatekeeping studies, there is also reason to believe
that editorial rank will have a lagged effect on what audiences choose to read. Following
this logic, we propose:
H3: Editorial judgments (e.g., placement of news stories on homepages) have a significant overall lagged effect on audience preferences (e.g., “most viewed” ranks), controlling for potential reciprocal effects. RQ2: What is the nature of the relationship between the lagged effects of editorial judgments on audience preferences?
Nevertheless, while causal relationships between audience preferences and
editorial judgments make theoretical sense in both directions, controlling for potential
reciprocal effects in analysis, we believe that the overall lagged effect of audience
preferences on editorial preferences will be stronger than the opposite lagged effect given
the popularization and promises of computer algorithms, coupled with most U.S.
commercial newsrooms’ pressure to drive online traffic and pageviews to please their
advertisers:
H4: The lagged effect of audience preferences on editorial judgments will be greater than the lagged effect of editorial judgments on audience preferences.
Method
This study performs secondary analysis on a set of data that were collected from
three New York-based news sites: NYTimes.com, NYPost.com, and NYDailyNews.com—
representing The New York Times, New York Post, and New York Daily News,
respectively. Data were gathered daily for two weeks, beginning June 1, 2010, and
ending June 14, 2010. On each day, data were collected at 9 a.m., 12 p.m., 3 p.m. and 6
18
p.m. Eastern Standard Time (EST), resulting in a total of 56 distinct collection shifts (14
days x 4 times per day) at each site and 1,550 total unique news stories. In each shift, the
top ten stories selected by journalists and by audiences were identified. “Top” stories in
terms of audience ranks (e.g., “most viewed” stories) and prominence of news story
placements were used as indicators of audience preferences and editorial judgments,
respectively.
Theoretical model explication. This study is essentially focused on estimating
two theoretical models at four distinct time points (9am, 12pm, 3pm and 6pm) over the
course of a day: The first looks at the lagged effect of editorial judgments at Time (X) on
audience preferences at Time (X+1), and the second looks at the lagged effect of
audience preferences at Time (X) on editorial judgments at Time (X+1). To bolster claim
about the direction of causality, this study also assesses potential reciprocal effects of
exogenous and endogenous variables where we account for the correlation between Y at
time 2 (12pm) and X at time 3 (3pm) in all estimated models. All of the statistical
analyses presented in this paper are done using maximum likelihood estimates, and under
the structural equation model framework.
Unit of analysis. Our unit of analysis is the story, which, following Boczkowski
and Peer (2011), we define as “text-based packages that include a headline” (p. 11).
Occasionally, a site would embed a sub-story within a particular story. At the point of
data analysis, we removed such sub-stories from the sample. The stories selected as the
top ten journalistic stories were the ten most prominently displayed, moving from below
the newspaper masthead and proceeding down the page. In the case of NYPost.com, only
five stories are listed; thus, at the data transformation stage, its rankings were multiplied
19
by two to enable a crude but comparable approximation. Stories identified as the top ten
audience preferences were located in a section halfway down the page, variously titled
“Most Clicked,” “Most Popular,” or “Most Read.”2
Variables. The following variables were used in this secondary analysis study:
editorial placement, audience rank, story ID. Editorial rank and audience rank, numbered
1 to 10, are simply the rank given to a particular story during a given collection period.
For a story displayed in only editorial and not audience rank (or vice versa), a value of 11
was given to the non-existent value as a way of flagging its rank as being less prominent
than any other possibility. New story IDs were assigned at 9 a.m. each day. These
numbers were then reverse coded during analysis to facilitate data interpretation, where
news stories that are not in the top 10 ranks are now given the value of 1, and new stories
that got ranked the most popular are given the value of 11. In other words, following this
new coding scheme, one unit increase in the reported coefficients is associated with the
effect of one rank increase (i.e., change from rank 5 to rank 4) in the exogenous variables
on the endogenous variables. Moreover, a news story-specific latent variable, α3, was
created to be included in all linear regression structural equation models.
Statistical models. This study estimates all the models using fixed effects
regression methods under the Structural Equation Modeling (SEM) framework where
simultaneous regression analyses were performed. H2 in particular enables interaction
2 Aside from “most viewed,” New York Times also has “most emailed,” “most blogged,” and “most searched,” and the New York Post has “most read” and “most commented.” On the other hand, the New York Daily News only has “most read,” “most discussed,” and “most emailed” categories, and we made the subjective decision that “most read” most closely resembles “most viewed” in the other two papers. 3 For models that estimate the lagged effect of editorial judgments on audience preferences, α is a latent variable measured by holding audience rank at time 2, 3, and 4 at 1, whereas for models that estimate the lagged effect of audience preferences on editorial judgments, α is a latent variable measured by holding editorial rank at time 2, 3, and 4 at 1.
20
with time to allow for observation of differential effects of the independent variables on
the dependent variables across the four observed time points; the rest of the models (H1,
RQ1, H3, RQ2, and H4) constrain the effects of independent variables on the dependent
variables to be the same across the four observed time points. Likelihood ratio
comparison test using differences in chi-square and degrees of freedoms4 were computed
to ensure that fixed effects analysis is not only theoretically, but also statistically, fit for
models estimated in this study. Data used in this analysis were transformed from “long
form” to “wide form” using the reshaping command in STATA for subsequent SEM
analysis in MPlus.
Fixed effects (FE) models. Fixed effects models are often used to analyze
longitudinal data with repeated measures on both independent and dependent variables.
FE models by default yield estimates that control for all stable characteristics of news
stories (whether observed or not), and allow for correlations between α, a news story-
specific latent variable, and each of the time-varying predictors. Moreover, fixed effects
model discard all between-news story variations in pursuit of “pure” results that are
approximately less biased than random effects models. Nevertheless, one of the
shortcomings of fixed effects models is that its pursuit of unbiased estimates is done at
the expense of model efficiency, which results in higher standard error estimates when
compared to random effects (Allison, 2005). Likelihood ratio comparison test using
differences in chi-square and degrees of freedom were computed, and the results
(p<0.001) suggest that in addition to its theoretical justifications, fixed effects are also
4 Note that this only tests the over-identifying restrictions, and is sensitive to sample size.
21
more appropriate, statistically, for models estimated in this study than that of random
effects models.
Benefits of estimating fixed effects models as a linear structural equation model
with a latent variable include its enabling (1) a likelihood ratio test for fixed versus
random effects to better assess the feasibility of each model, (2) estimations of models
with reciprocal effects among endogenous and exogenous variables, (3) assessments of
models with latent variables that have multiple indicators, and (4) evaluations of panel
data where some variables are believed to have lagged or reciprocal effects on each other
(Allison, 2009).
Results
Lagged effects of audience preferences on editorial judgments
Supportive of H1, there exists a significant overall lagged effect of audience
preferences on editorial judgments (p<0.001), controlling for effects of earlier editorial
judgments on later editorial judgments, as well as potential reciprocal effects between
editorial judgments and subsequent audience preferences. Also, there is no evidence (n.s.)
for a reciprocal effect between audience preferences at time 2 on editorial judgments at
time 3 in this estimated model.
Goodness of Fit tests for H1 & RQ1. Converging evidence of a variety of
Goodness of Fit (GOF) tests suggests that the estimated model is a relatively good fit for
the data: The overall chi-square test of model fit suggests that the estimated model is not
a just-identified model, X2 (7) = 64.77, p<.01, which is consistent with the theory-driven
constraints imposed on this model. Ideally, we would want the chi-square test for overall
model fit to be non-significant for non-saturated models like ours; however, given this
22
study’s large sample size and chi-square tests’ sensitivity to sample size, the statistical
significance found in this particular GOF test is not surprising. On the other hand, the
relatively small chi-square test value offers support for the fitness of our proposed model.
The Root Mean Square Error of Approximation (RMSEA) is .07. The Comparative Fit
Index (CFI) is .98. The Tucker-Lewis Fit Index (TLI) is .94. The Standardized Root
Mean Square Residual (SRMR) is .04.
In response to RQ1, regardless of which model one focuses on, audience
preferences appear to have an overall negative lagged influence on editorial judgments in
that a one-rank advancement in audience preferences for a news story at Time (X) is
associated with such story’s subsequently decline (B= -.15, p<.001) in editorial
placement at Time (X+1), controlling for everything else.
Supportive of H2, lagged effects of audience preferences on editorial judgments
do strengthen steadily throughout the four recorded time points on an average day when
comparing the percentage of variance explained (R2) by audience preference throughout
the day, controlling for effects of earlier audience preferences on later audience
preferences, as well as potential reciprocal effects between audience preferences and
subsequent editorial judgments. Particularly, audience preference explains for 4% of the
variance at 9am, 15% of the variance at 12pm, 14% of the variance at 3pm, and 25% of
the variance at 6pm. All four R2 values are statistically significant at p<.001.
Goodness of Fit tests for H2. A variety of Goodness of Fit (GOF) tests offers
converging support that the estimated model is a relatively good fit for the data: The
overall chi-square test of model fit suggests that the estimated model is not a just-
identified model, X2 (5) = 61.99, p<.01, which is consistent with the theory-driven
23
constraints imposed on this model. As aforementioned, given this study’s large sample
size and chi-square tests’ sensitivity to sample size, the statistical significance found in
this particular GOF test is not surprising. On the other hand, the small chi-square value
offers support for the fitness of this particular model. The Root Mean Square Error of
Approximation (RMSEA) is .09. The Comparative Fit Index (CFI) is .98. The Tucker-
Lewis Fit Index (TLI) is .92. The Standardized Root Mean Square Residual (SRMR) is
.04.
Lagged effects of editorial judgments on audience preferences
Unsupportive of H3, editorial judgments of news stories has no overall lagged
effects on audience preferences (n.s.).
Goodness of Fit tests for H3. The overall chi-square test of model fit suggests
that the estimated model is not a just-identified model, X2 (7) = 87.26, p <.01, which is
consistent with the theory-driven constraints imposed on this model. As aforementioned,
significant chi-square test of overall model fit is not surprising given the large sample
size used in this study. On the other hand, the relatively small chi-square value lends
support to the fitness of the proposed model. The Root Mean Square Error of
Approximation (RMSEA) is .09. The Comparative Fit Index (CFI) is .98. The Tucker-
Lewis Fit Index (TLI) is .94. The Standardized Root Mean Square Residual (SRMR) is
.05.
In response to RQ2, there is no overall lagged effect of editorial judgment on
audience preferences.
24
Supportive of H4, comparing results found in H1 and H3, lagged effect of
audience preferences on editorial judgments is greater than the lagged effect of editorial
judgments on audience preferences.
Discussion and Conclusion
In the beginning of the paper we raised the question about the directionality of
influence between editorial news judgments and audience preferences, and the short
answer is that the causal influence is not uni-directional. This study suggests that data
aggregation of audience preferences plays an intricate and dynamic role in influencing
whether and how online newsroom editors decide to feature certain news stories over
others at multiple time points a day. While penny press editors in the 1830s might have
looked over the shoulders of newspaper readers on city streets to help them make
decisions about future story coverage and placement, this study provides a quantitative
indication that online newsroom editors today are relying more and more on digital
tracking tools to understand audience desires in order to maximize their presentation of
content that audiences will find favorable.
To summarize our key findings, we found that, (1) controlling for potential
reciprocal effects, audience preferences affect subsequent editorial placement of news
stories, based on a time-lagged analysis. Moreover, (2) the strength of this effect on
editorial judgments appears to intensify during the course of the day. Meanwhile, looking
the other direction, (3) there is no overall lagged effect of editorial judgments on
audience preferences—yet subsequent analysis suggests that this finding may be
explained by the fluid and conflicting nature of the lagged effects we observed during the
25
course of the day. Finally, (4) the lagged effect of audience preferences on editorial
judgments is stronger than the inverse relationship.
That audience preferences would appear to have greater impact, at least in this
initial study, casts in a new light some baseline assumptions about the agenda-setting and
gatekeeping functions of news editors. This is particularly so given that such assumptions
were established through research conducted long before the widespread newsroom use
of audience information systems and sophisticated computer algorithms. The literature
has shown that journalists “have long-established processes for assessing information and
determining what they believe should go into the news product—and are extremely
reluctant to relinquish control over those decisions, despite the greatly increased visibility
of user activity on media-affiliated websites” (Singer, 2011, p. 630). And yet an emerging
body of ethnographic research (notably Anderson, 2011a,b) also suggests that metrics
might be influencing news decisions—at least to some degree and contingent on
management expectations—as journalists become increasingly aware of traffic data
circulated within the newsroom and of “most viewed” lists prominent on homepages,
these artifacts representing both internal and external manifestations of audience desires.
Even still, other research (notably Boczkowski et al., 2011) has indicated that the gap
between journalists and audiences remains quite large, each group wanting something
different from the news experience online.
Amid somewhat conflicting notions about convergence and divergence of
editorial judgment and audience preferences, this study is perhaps the first of its kind to
use quantitative methods to identify an actual time-lagged influence one way or
another—to tell us something about the relative “power” in this relationship. Thus, our
26
most striking finding is that editorial judgments of what is “important” do not necessarily
transfer to the minds of audiences, if we take “most viewed” as proximal indication of
issue salience for audiences (i.e., the “agenda of the audience”). In other words, the
influence of audiences upon journalists was found to be significant while the inverse
relationship was not. This would suggest that, all things being equal, journalists appear to
be more aware of and responsive to audience desires than the other way around. This
supports the conclusions of Boczkowski and his colleagues (see Boczkowski, 2010;
Boczkowski et al., 2011; Boczkowski & Peer, 2011) that there is a substantial disparity
between what audiences want to read and what journalists feel that they should read. But
this paper goes further in exploring the relative differences of influence at work in this
gap, revealing that, over the course of a typical day, audiences are disregarding the news
judgments of homepage editors even while journalists—at least to some degree—appear
to be influenced by what they know about audience tastes.
Yet, it is likewise important to note that further research is needed to understand
the underlying dynamics; for even when lagged effects were found, after relaxing time
constraints, the strength of the effect was minimal. Moreover, regarding these lagged
effects, we first found that audience preferences appear to have an overall negative
lagged influence on editorial judgments (see RQ1). On its face, this seems a bit
counterintuitive: If journalists indeed are being influenced by audience interests, should
that relationship not likely be positive—i.e., editors promoting already popular stories to
the top of the page, to capitalize on that popularity? Perhaps editors are bench-marking
against audience preferences by actively pushing such stories down the homepage,
presuming that such stories already have prominent placement merely by being in the
27
“most viewed” list on the homepage? Additional research, linking both qualitative
examinations of newsroom decision-making and quantitative analyses such as this, is
needed to untangle the precise nature of this lagged influence. In any case, given online
news media’s emphasis on speed and immediate updates (Lee, 2011), this negative
coefficient makes sense in that most online newspapers routinely and frequently update
their homepages throughout the day as new events occur or existing happenings evolve.
Naturally, “newer” news stories will be made more prominent on homepages.
Next, regarding the lagged effects of editorial judgments on audience preferences
and vice versa, the juxtaposition of findings from RQ1 and RQ2 points to the ostensible
rise of audience power, as news organizations increasingly monitor online visitor traffic
patterns and appear to take some cues from this data in story placement on newspaper
homepages. Specifically, the findings suggest that whereas online editors are mindful of
what news audiences are clicking on, news audiences are not paying as much attention to
what news editors think they should read. Instead, if we take “most viewed” ranking as
an indication of news story popularity, this study reinforces Boczkowski and colleagues’
finding that editors and audiences want different things from the news—and it further
adds to the literature by indicating that, at the crossroad of this taste disparity, news
audiences have the upper-hand.
There are additional limitations to this study. One is our reliance on secondary
data. Ideally, this study would have included a wider range of newspapers for
examination from different geographical markets or newsroom practices to supplement
the external validity of our findings. However, as a secondary data analysis, this study
bases its analysis on three New York-based newspapers collected in the data set instead,
28
and thus future studies are encouraged to test our proposed models using other
newspapers to help validate the generalizability of our findings to newspapers in other
markets and with different newsroom routines. While time-lagged analysis under the
Structural Equation Modeling framework is known for its strength in enabling causal
testing5 in ways that come close to being comparable to the internal validity strengths of
true experimental designs, it is not without limitations. One limitation surrounds the
subjective justification of the “right time lag” in time-lagged analysis, as it is possible
that a 2-hour lagged effect is different from a 4-hour lagged effect, and the best
researchers can do is to ensure that the estimated time lag intervals make the best
theoretical and practical sense possible. In this study, we interpreted the best time-lag for
analysis to be four hours, although future studies are encouraged to embed different time
lags in their study design as to experiment whether different time lags contribute to
different findings or not. Finally, another limitation is our assumption that “most viewed”
equates “audience preferences.” Some may argue they are not the same thing, yet if
media choices are assumed to be audience driven—as in the classic approach of uses and
gratifications—it is reasonable to use “most viewed” as the best available proxy for
measuring audience preferences.
Ultimately, and despite the limitations of this research, we argue that the central
finding of this paper—that audience preferences have a lagged effect on editorial
judgments, all things held constant—is a provocative one that deserves attention for how
it may spur additional work to uncover the complex relationship between audience data
and news work, against the theoretical backdrop of agenda-setting and gatekeeping 5 SEM facilitates causal testing through its control of potential reciprocal effects between the independent and dependent variables, reliance on temporal precedence, and assessment of model fitness via GOF tests.
29
theories of journalism. Even as journalists seek to preserve their autonomy and guard
against becoming enslaved to audiences’ desires for “softer” news, it is clear that
audience metrics are here to stay—and, going forward, are likely to become only more
embedded in the process of news judgment. The challenge for researchers, and one that
we have sought to take up in this study, is to move beyond self-reports of journalistic
perception and behavior, and instead use quantitative methods that reveal a more precise,
longitudinal rendering of the relationship between audience and editorial preferences.
30
References
Anderson, C. W. (2011a). Between creative and quantified audiences: Web metrics and changing patterns of newswork in local US newsrooms. Journalism, 12(5), 550. doi:10.1177/1464884911402451
Anderson, C. W. (2011b). Deliberative, agonistic, and algorithmic audiences: Journalism's vision of its public in an age of audience transparency. International Journal of Communication, 5(0), 529-547.
Ang, I. (1991) Desperately Seeking the Audience. London: Routledge. Allison, P. D. (2009). Fixed Effects Regression Models. California: SAGE Publications. Allison, P. D. (2005). Fixed Effects Regression Methods for Longitudinal Data Using
SAS. North Carolina: SAS Publishing. Barzilai-Nahon, K. (2009). Gatekeeping: A critical review. Annual Review of Information
Science and Technology, 43(1), 1-79. doi:10.1002/aris.2009.1440430117 Berger, J., & Milkman, K. L. (2012). What makes online content viral? Journal of
Marketing Research. Bermejo, F. (2009). Audience manufacture in historical perspective: From broadcasting
to google. New Media & Society, 11(1-2), 133. doi:10.1177/1461444808099579 Boczkowski, P. J. (2010). News at work: Imitation in an age of information abundance.
Chicago: University of Chicago Press. Boczkowski, P. J., & Mitchelstein, E. (2012). How users take advantage of different
forms of interactivity on online news sites: Clicking, e-mailing, and commenting. Human Communication Research, 38(1), 1-22. doi:10.1111/j.1468-2958.2011.01418.x
Boczkowski, P. J., & Peer, L. (2011). The choice gap: The divergent online news preferences of journalists and consumers. Journal of Communication, 61(5), 857-876. doi:10.1111/j.1460-2466.2011.01582.x
Boczkowski, P. J., Mitchelstein, E., & Walter, M. (2011). Convergence across divergence: Understanding the gap in the online news choices of journalists and consumers in Western Europe and latin america. Communication Research, 38(3), 376-396. doi:10.1177/0093650210384989
Bogart, L. (1957). Opinion research and marketing. Public Opinion Quarterly, 21(1), 129-140. doi:10.1086/266692
Bruns, A. (2008). Blogs, Wikipedia, Second Life, and beyond: From production to produsage. New York: Peter Lang. Retrieved from http://snurb.info/files/Produsage%20-%20Introduction.pdf
Chaffee, S. H., & Metzger, M. J. (2001). The end of mass communication? Mass Communication and Society, 4(4), 365-379.
Cohen, E. L. (2002). Online journalism as market-driven journalism. Journal of Broadcasting & Electronic Media, 46(4), 532-548.
Dick, M. (2011). Search engine optimisation in UK news production. Journalism Practice, 5(4), 462-477. doi:10.1080/17512786.2010.551020
Domingo, D., Quandt, T., Heinonen, A., Paulussen, S., Singer, J. B., & Vujnovic, M. (2008). Participatory journalism practices in the media and beyond: An international comparative study of initiatives in online newspapers. Journalism Practice, 2(3), 326-342. doi:10.1080/17512780802281065
31
Ettema, J. S., & Whitney, D. C. (1994). The money arrow: An introduction to audiencemaking. In J. S. Ettema & D. C. Whitney (Eds.), Audiencemaking: How the media create the audience. (pp. 1-18). New York: Sage.
Fishman, M. (1980). Manufacturing the news. Austin: University of Texas Press. Gans, H. J. (2004). Deciding what's news: A study of CBS Evening News, NBC Nightly
News, Newsweek, and TIME. New York: Pantheon Books. Gillmor, D. (2004). We the media: Grassroots journalism by the people, for the people.
Sebastopol, CA: O'Reilly Media. Graves, L., & Kelly, J. (2010). Confusion online: Faulty metrics & the future of digital
journalism. New York: Columbia University Graduate School of Journalism. Retrieved from http://www.journalism.columbia.edu/page/633/437
Harrison, T. M., & Barthel, B. (2009). Wielding new media in Web 2.0: Exploring the history of engagement with the collaborative construction of media products. New Media & Society, 11(1-2), 155-178. doi:10.1177/1461444808099580
Hermida, A. (2010). Twittering the news: The emergence of ambient journalism. Journalism Practice, 4(3), 297-308. doi:10.1080/17512781003640703
Jenkins, H., Ford, S., & Green, J. (2012). Spreadable media: Creating value and meaning in a networked culture. New York: NYU Press.
Jönsson, A. M., & Örnebring, H. (2011). User-Generated content and the news: Empowerment of citizens or interactive illusion? Journalism Practice, 5(2), 127-144.
Lee, A. M. (2011). Is the medium the message? Predicting popularity of top news sites with medium-specific features. #ISOJ,1(2),153-189.
Lee, M. (2011). Google ads and the blindspot debate. Media, Culture & Society, 33(3), 433-447. doi:10.1177/0163443710394902
Lewis, S. C., Kaufhold, K., & Lasorsa, D. L. (2010). Thinking about citizen journalism: The philosophical and practical challenges of user-generated content for community newspapers. Journalism Practice, 4(2), 163-179. doi:10.1080/14616700903156919
Loosen, W., & Schmidt, J. (2012). (Re-)discovering the audience: The relationship between journalism and audience in networked digital media. Forthcoming in Information, Communication & Society.
Lowrey, W. (2009). Institutional roadblocks: Assessing journalism's response to changing audiences. In Z. Papacharissi (Ed.), Journalism and citizenship: New agendas. (pp. 44-67). Mahwah, NJ: Lawrence Erlbaum Associates (LEA/Taylor & Francis).
MacGregor, P. (2007). Tracking the online audience: Metric data start a subtle revolution. Journalism Studies, 8(2), 280-298. doi:10.1080/14616700601148879
McCombs, M. (2005). A look at agenda-setting: Past, present and future. Journalism Studies, 6(4), 543-557. doi:10.1080/14616700500250438
McManus, J. H. (1994). Market-Driven journalism: Let the citizen beware? Thousand Oaks, CA: Sage.
Mersey, R. D. (2010). Can journalism be saved?: Rediscovering America's appetite for news. Santa Barbara, CA: Praeger.
Mitchelstein, E., & Boczkowski, P. J. (2009). Between tradition and change: A review of recent research on online news production. Journalism, 10(5), 562-586.
32
doi:10.1177/1464884909106533 Mitchelstein, E., & Boczkowski, P. J. (2010). Online news consumption research: An
assessment of past work and an agenda for the future. New Media & Society, 12(7), 1085-1102. doi:10.1177/1461444809350193
Mortensen, M. (2011). When citizen photojournalism sets the news agenda: Neda Agha Soltan as a Web 2.0 icon of post-election unrest in Iran. Global Media and Communication, 7(1), 4. doi:10.1177/1742766510397936
Napoli, P. M. (2010). Audience evolution: New technologies and the transformation of media audiences. New York: Columbia University Press.
Picone, I. (2011). Produsage as a form of self-publication. A qualitative study of casual news produsage. New Review of Hypermedia and Multimedia, 17(1), 99-120. doi:10.1080/13614568.2011.552643
Postigo, H. (2011). Questioning the web 2.0 discourse: Social roles, production, values, and the case of the human rights portal. The Information Society: An International Journal, 27(3), 181-193. doi:10.1080/01972243.2011.566759
Reinemann, C., Stanyer, J., Scherr, S., & Legnante, G. (2012). Hard and soft news: A review of concepts, operationalizations and key findings. Journalism, 13(2), 221-239. doi:10.1177/1464884911427803
Robinson, S. (2011a). Convergence crises: News work and news space in the digitally transforming newsroom. Journal of Communication, 61(6), 1122-1141. doi:10.1111/j.1460-2466.2011.01603.x
Robinson, S. (2011b). Journalism as process: The labor implications of participatory content in news organization. Journalism & Communication Monographs, 13(3), 138-210.
Rosen, J. (2006). The people formerly known as the audience. Pressthink [Web page]. Retrieved from http://journalism.nyu.edu/pubzone/weblogs/pressthink/2006/06/27/ppl_frmr.html
Schlesinger, P. (1987). Putting “reality” together: BBC news. London: Methuen. Schudson, M. (1992). The sociology of news. In J.Curran & M.Gurevitch (Eds.), Mass
Media and Society (pp. 41-59). London: Edward Arnold. Shirky, C. (2008). Here comes everybody: The power of organizing without
organizations. New York: Penguin Press. Shoemaker, P. J., & Reese, S. D. (1996). Mediating the message: Theories of influences
on mass media content. White Plains, NY: Longman. Shoemaker, P. J., & Vos, T. P. (2009). Gatekeeping theory. New York: Routledge. Singer, J. B. (2011). Community service: Editor pride and user preference on local
newspaper websites. Journalism Practice, 5(6), 623–642. doi:10.1080/17512786.2011.60193
Singer, J. B., Domingo, D., Heinonen, A., Hermida, A., Paulussen, S., Quandt, T., . . . Vujnovic, M. (2011). Participatory journalism: Guarding open gates at online newspapers. Malden, MA: Wiley-Blackwell.
Smythe, D. (1977) Communications: Blindspot of Western Marxism. Canadian Journal of Political and Social Theory, 1(3): 1-27.
Thorson, E. (2008). Changing patterns of news consumption and participation: News recommendation engines. Information, Communication & Society, 11(4), 473-489. doi:10.1080/13691180801999027
33
Thurman, N. (2011). Making ‘the daily me’: Technology, economics and habit in the mainstream assimilation of personalized news. Journalism, 12(4), 395-415. doi:10.1177/1464884910388228
Tuchman, G. (1978). Making news: A study in the construction of reality. New York: Free Press.
Turow, J. (2005). Audience construction and culture production: Marketing surveillance in the digital age. The ANNALS of the American Academy of Political and Social Science, 597(1), 103-121. doi:10.1177/0002716204270469
Weaver, D. H., & Wilhoit, G. C. (1996). The American Journalist in the 1990s: U. S. News People at the End of an Era. Mahwah, NJ: Lawrence Erlbaum Associates.